Qurie is built with 5 key design elements in mind:

(1) Qurie provides an end-to-end solution that integrates multiple segments of a researcher’s workflow, eliminating the encumbrance of a poorly integrated workflow.

(2) Fast and accurate AI analysis frees researchers to concentrate on complex objectives that require human expertise, rather than being tied up in the tedious and time-consuming task of counting mitotic figures.

(3) AI performs the heavy lifting of locating mitotic figures, which can then be easily verified by researchers. This hybrid model of human-AI analysis combines the speed of AI analysis with the reliability of human histopathologists for the dual benefits of efficiency and accuracy. 

(4) Analyses are transformed into aggregated insights which summarise the most salient information for easy reference and digestibility.

(5) Qurie’s digital slideviewer pushes the practice of squinting into microscope lenses into obscurity with a smooth and seamless slide viewing experience, reaching magnification levels of up to 40x with minimal latency.

Qurie delivers accurate, histopathologist-approved AI analysis in less than 8 minutes per slide - several times the speed of human histopathologists. Qurie’s tumour segmentation model returns an intersection-over-union value of 0.72 - demonstrating an accuracy comparable to human analysis while offering more precise boundaries than humans could manually annotate.

As for usability, feedback from user testing strongly indicated that the platform was easy and intuitive to use. 80% of users found the platform very easy to use and 70% of users felt strongly that they wouldn't require the help of a technical person to operate the platform. Qurie was also praised for its minimalist design and ease of navigation.

Histopathology involves complex tasks, such as evaluating the mitotic activity index (MAI), an important indicator used to determine the malignancy of tumours. This entails: 

(1) identifying tumour areas

(2) counting the number of replicating cells, known as mitotic figures. 

Mitotic figures are 200x thinner than human hair, and even with the help of a microscope, identifying mitotic figures is practically impossible for a layperson. Even trained histopathologists take around 30 minutes to analyse just 10-20% of a slide - a tedious and time-consuming process.

The AI behind Qurie consists of 2 AI models:

(1) a segmentation model, used to identify tumor areas

(2) an object detection model, used to detect mitotic figures

AI models are taught how to complete these tasks through training - a process in which models are fed image data along with labels - and gradually learn how to segment tumours as well as to detect mitotic figures.Once the models have been trained, Qurie uses them to produce predictions on unseen images, a process known as inference.

For each of the 2 tasks, multiple models and backbones were experimented with and various optimisations implemented, to create the fast and accurate AI that powers Qurie.

We began this project with a mission - to use design and AI to augment cancer research. By using AI to automate the painstaking process of histopathology analysis, Qurie is a game-changing tool which we hope will support researchers in the fight against cancer.

Qurie will be deployed by the team’s industry partner, Qritive, to various organisations.  Qurie’s modular design facilitates the easy addition of new AI modules and features, allowing Qritive to tailor the platform to fit the needs of different clients. The sky is the limit for AI within the field of cancer research, and Qritive can build on Qurie to push its boundaries.